The Flash Flip: Google’s New Gemini 3 Flash Topples GPT-5.2 in Key AI Benchmarks
The struggle for leadership in AI is increasingly shifting from research labs to mass-market products. Just one day after the release of OpenAI’s GPT-Image-1.5, Google unveiled Gemini 3 Flash and immediately began rolling it out as the default model across its consumer applications, as well as within Search’s AI Mode. The new model replaces Gemini 2.5 Flash, which debuted barely six months ago.
Google says Gemini 3 Flash delivers marked gains in reasoning, multimodality, and efficiency while maintaining low latency. Early evaluations suggest the model is closing in on so-called “frontier” systems from both Google and OpenAI. In the Humanity’s Last Exam benchmark, Gemini 3 Flash scored 33.7% without tool use, compared with 37.5% for Gemini 3 Pro, 11% for Gemini 2.5 Flash, and 34.5% for the recently released OpenAI GPT-5.2. In the multimodal MMMU-Pro test, Gemini 3 Flash achieved 81.2%, which Google claims is the strongest result among competitors in this category.
Despite these performance gains, Google continues to position Flash as a “speed-first” model. The company states that it outperforms Gemini 2.5 Pro while running at roughly three times the speed, and that in reasoning tasks it consumes about 30% fewer tokens on average—potentially translating into lower operational costs.
For everyday users, Gemini 3 Flash has already become the default option in the Gemini app worldwide, while Gemini 3 Pro remains available via the model selector for more demanding tasks in mathematics and programming. Google is also leaning heavily into multimodal use cases: uploading short videos for guidance and coaching, submitting sketches for recognition, and working with audio for analysis or quiz creation. The company notes that the model more frequently offers “visual answers,” including images and tables, and can even help assemble simple application prototypes directly within the Gemini interface from natural-language prompts.
In Search, Google is simultaneously broadening access to advanced capabilities. Gemini 3 Pro is now available to all users in the United States, and the company has expanded deployment of its in-house image generation model, Nano Banana Pro, within search results. On the enterprise side, Google reports adoption of Gemini 3 Flash by companies such as JetBrains, Figma, Cursor, Harvey, and Latitude, with the model accessible through Vertex AI and Gemini Enterprise. Developers are promised preview access via API, as well as integration into Antigravity, a coding tool Google released last month.
Google has set API pricing at $0.50 per million input tokens and $3.00 per million output tokens—slightly higher than Gemini 2.5 Flash, a difference the company attributes to improved quality. Google also claims that, following the launch of the Gemini 3 family, its APIs are now processing more than one trillion tokens per day. Against this backdrop, competition with OpenAI is intensifying: reports earlier in December referenced an internal “Code Red” memo from Sam Altman after a dip in ChatGPT traffic amid Google’s rising share, followed by OpenAI’s release of GPT-5.2 and a new image generation model, alongside claims that message volume in ChatGPT has grown eightfold compared with November 2024.
Support Our Threat Intelligence
If you find our technology report and cybersecurity news helpful, consider supporting our work.
Buy Me a Coffee
PayPal
